arch/powerpc/lib/xor_vmx.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License as published by
   4  * the Free Software Foundation; either version 2 of the License, or
   5  * (at your option) any later version.
   6  *
   7  * This program is distributed in the hope that it will be useful,
   8  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  10  * GNU General Public License for more details.
  11  *
  12  * You should have received a copy of the GNU General Public License
  13  * along with this program; if not, write to the Free Software
  14  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  15  *
  16  * Copyright (C) IBM Corporation, 2012
  17  *
  18  * Author: Anton Blanchard <anton@au.ibm.com>
  19  */
  20
  21 /*
  22  * Sparse (as at v0.5.0) gets very, very confused by this file.
  23  * Make it a bit simpler for it.
  24  */
  25 #if !defined(__CHECKER__)
  26 #include <altivec.h>
  27 #else
  28 #define vec_xor(a, b) a ^ b
  29 #define vector __attribute__((vector_size(16)))
  30 #endif
  31
  32 #include <linux/preempt.h>
  33 #include <linux/export.h>
  34 #include <linux/sched.h>
  35 #include <asm/switch_to.h>
  36
  37 typedef vector signed char unative_t;
  38
  39 #define DEFINE(V)                               \
  40         unative_t *V = (unative_t *)V##_in;     \
  41         unative_t V##_0, V##_1, V##_2, V##_3
  42
  43 #define LOAD(V)                 \
  44         do {                    \
  45                 V##_0 = V[0];   \
  46                 V##_1 = V[1];   \
  47                 V##_2 = V[2];   \
  48                 V##_3 = V[3];   \
  49         } while (0)
  50
  51 #define STORE(V)                \
  52         do {                    \
  53                 V[0] = V##_0;   \
  54                 V[1] = V##_1;   \
  55                 V[2] = V##_2;   \
  56                 V[3] = V##_3;   \
  57         } while (0)
  58
  59 #define XOR(V1, V2)                                     \
  60         do {                                            \
  61                 V1##_0 = vec_xor(V1##_0, V2##_0);       \
  62                 V1##_1 = vec_xor(V1##_1, V2##_1);       \
  63                 V1##_2 = vec_xor(V1##_2, V2##_2);       \
  64                 V1##_3 = vec_xor(V1##_3, V2##_3);       \
  65         } while (0)
  66
  67 void xor_altivec_2(unsigned long bytes, unsigned long *v1_in,
  68                    unsigned long *v2_in)
  69 {
  70         DEFINE(v1);
  71         DEFINE(v2);
  72         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
  73
  74         preempt_disable();
  75         enable_kernel_altivec();
  76
  77         do {
  78                 LOAD(v1);
  79                 LOAD(v2);
  80                 XOR(v1, v2);
  81                 STORE(v1);
  82
  83                 v1 += 4;
  84                 v2 += 4;
  85         } while (--lines > 0);
  86
  87         disable_kernel_altivec();
  88         preempt_enable();
  89 }
  90 EXPORT_SYMBOL(xor_altivec_2);
  91
  92 void xor_altivec_3(unsigned long bytes, unsigned long *v1_in,
  93                    unsigned long *v2_in, unsigned long *v3_in)
  94 {
  95         DEFINE(v1);
  96         DEFINE(v2);
  97         DEFINE(v3);
  98         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
  99
 100         preempt_disable();
 101         enable_kernel_altivec();
 102
 103         do {
 104                 LOAD(v1);
 105                 LOAD(v2);
 106                 LOAD(v3);
 107                 XOR(v1, v2);
 108                 XOR(v1, v3);
 109                 STORE(v1);
 110
 111                 v1 += 4;
 112                 v2 += 4;
 113                 v3 += 4;
 114         } while (--lines > 0);
 115
 116         disable_kernel_altivec();
 117         preempt_enable();
 118 }
 119 EXPORT_SYMBOL(xor_altivec_3);
 120
 121 void xor_altivec_4(unsigned long bytes, unsigned long *v1_in,
 122                    unsigned long *v2_in, unsigned long *v3_in,
 123                    unsigned long *v4_in)
 124 {
 125         DEFINE(v1);
 126         DEFINE(v2);
 127         DEFINE(v3);
 128         DEFINE(v4);
 129         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
 130
 131         preempt_disable();
 132         enable_kernel_altivec();
 133
 134         do {
 135                 LOAD(v1);
 136                 LOAD(v2);
 137                 LOAD(v3);
 138                 LOAD(v4);
 139                 XOR(v1, v2);
 140                 XOR(v3, v4);
 141                 XOR(v1, v3);
 142                 STORE(v1);
 143
 144                 v1 += 4;
 145                 v2 += 4;
 146                 v3 += 4;
 147                 v4 += 4;
 148         } while (--lines > 0);
 149
 150         disable_kernel_altivec();
 151         preempt_enable();
 152 }
 153 EXPORT_SYMBOL(xor_altivec_4);
 154
 155 void xor_altivec_5(unsigned long bytes, unsigned long *v1_in,
 156                    unsigned long *v2_in, unsigned long *v3_in,
 157                    unsigned long *v4_in, unsigned long *v5_in)
 158 {
 159         DEFINE(v1);
 160         DEFINE(v2);
 161         DEFINE(v3);
 162         DEFINE(v4);
 163         DEFINE(v5);
 164         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
 165
 166         preempt_disable();
 167         enable_kernel_altivec();
 168
 169         do {
 170                 LOAD(v1);
 171                 LOAD(v2);
 172                 LOAD(v3);
 173                 LOAD(v4);
 174                 LOAD(v5);
 175                 XOR(v1, v2);
 176                 XOR(v3, v4);
 177                 XOR(v1, v5);
 178                 XOR(v1, v3);
 179                 STORE(v1);
 180
 181                 v1 += 4;
 182                 v2 += 4;
 183                 v3 += 4;
 184                 v4 += 4;
 185                 v5 += 4;
 186         } while (--lines > 0);
 187
 188         disable_kernel_altivec();
 189         preempt_enable();
 190 }
 191 EXPORT_SYMBOL(xor_altivec_5);